Home
Archives
Archives before June 13, 2001
RSS Feed
Adaptive
Path (my company!)
About
peterme
Coordinates
Most of the Time
Oakland, CA
Interests
Current
American history around the time of the Revolution, figuring out how to marry top-down task-based information architecture processes with bottom-up document-based ones, finding a good dentist in San Francisco Oakland
Perennial
Designing
the user experience (interaction design, information architecture, user
research, etc.), cognitive science, ice cream, films and film theory,
girls, commuter bicycling, coffee, travel, theoretical physics for laypeople,
single malt scotch, fresh salmon nigiri, hanging out, comics formalism,
applied complexity theory, Krispy Kreme donuts.
surf
Click
to see where I wander.
Wish
list
Show
me you love me by
buying
me things.
Spyonme
Track updates of
this page with Spyonit. Clickee
here.
Essays
[Editor's note: peterme.com
began as a site of self-published essays, a la Stating
The Obvious. This evolved (or devolved) towards link lists and shorter
thoughtpieces. These essays are getting a tad old, but have some good
ideas.]
Reader Favorites
Interface
Design Recommended Reading List
Whose
"My" Is It Anyway?
Frames:
Information Vs. Application
Subjects
Interface Design
Web Development
Movie Reviews
Travel
| Innovation in Classification. Posted on 09/23/2001. |
|
The second thread involves faceted classification, one of the most powerful, yet least understood, methods of organizing information. Most folks, when thinking about organizing objects or information, immediately think of a hierarchical, or taxonomic, organization; a top-down structure, where you start with a number of broad categories that get ever more detailed, until you arrive at the object. In such structures, each object has a single home, and typically, one path to get there--this is how things are organized in "the real world", where each item can only be in one place. Oftentimes, when thinking of organizing information, a hierarchy is where people begin (think Yahoo!). Faceted classification, on the other hand, is a bottom-up scheme. Here, each object is tagged with a certain set of attributes and values (these are the facets), and the organization of these objects emerges from this classification, and how a user chooses to access them. Toys, for example, lend themselves to a faceted classification, with the facets being things like, "Suitable Age," "Price," "Subject Type," "Brand," and even "Character" (like Barbie or Elmo). Someone might be price conscious, and want to start there; another knows that the child in question loves science toys, and wants to begin with that. Faceted classification allows for exploration directed by the user, where a large dataset is progressively filtered through the user's various choices, until arriving at a manageable set that meet the users' basic criteria. Instead of sifting through a pre-determined hierarchy, the items are organized on-the-fly, based on their inherent qualities. Now, faceted classification isn't inherently innovative. In fact, objects tend to have a fixed set of facets by which they are organized. Where innovation comes is through user research that listens to how the users/customers/audience think about and approach a task, and providing tools to allow them to approach it meaningfully. Wine is a much-fussed-over subject that, over the years, has developed a language and organization of it's own, an organization that happens to be faceted. Wines are typically organized by color, or varietal, or region, and occasionally by price. Online wine merchants, like Wine.com (now eVineyard), have exploited this, allowing their visitors to shop for wine in these time-tested fashions.
Anything that is "time-tested" is ripe for innovation. I know I've been utterly at a loss when shopping at a store or online for wine, because, I don't understand the differences in region and variety, and, frankly, I don't care. I have a sense of what I like, or I know what I'll be drinking it with, but the store layouts are of no help. One smart entrepreneur realized this, and opened up a wine store, Best Cellars, that eschews the arcana of wine connoisseurship in favor of a classification scheme better suited to the average drinker:
These, I understand! I know I tend to like "medium-to-full bodied" red wines, and that's easy to find here. The proprietor ditched conventional wisdom, utilizing terminology they knew their users would better understand. In fact, the whole store is premised on this innovative nomenclature scheme, and it's been frightfully successful. User-centered approaches engender innovation by encouraging us to truly listen to and understand our audience, in order to best serve their unmet needs. In the same way that the focus of Design is shifting away from products (the thing in itself) and towards processes (things in context, in relation to people and the environment), innovation, which once meant "technological innovation," will increasingly mean procedural innovation--a natural outcome of user-centered methodology. For folks interested in more advanced takes: 32 comments so far. Add a comment. Previous entry: "Beware: Time Waster Ahead." COMMENT #1 The links are much appreciated, too. :) Suzanne
COMMENT #2 these two approaches are by no means exclusive, and this shouldn't be an "either or" discussion. any faceted data set would support a number of different user views, including a hierarchical view of topics. the wine shopping examples you've raised are in fact both examples of both faceted _and_ hierarchical views; i would say these are both examples of tailoring a classification system to a particular audience (in this case, it seems the wine menus were created for expert vs. novice audiences). faceted classification, while it's an important concept, isn' t a particularly new idea (though it's probably new to a lot of Web folks). librarians and DBAs have been trafficking in faceted classification systems for decades (though they may not all use that term), ever since Ranganathan. all the major library classification systems are built around this idea. and relational database architects regularly apply myriad descriptive categories to all kinds of data (though they wouldn't use the term "facets"). anyway, good topic - and thanks for the meaty links (nothing like reading a few library classification white papers with your morning coffee...). - alex
COMMENT #3 But to Alex's point. He is right in saying that the concept of facet analysis is not a new alternative to hierarchies of subject categories, although putting its application and use in the hands of users might be. The idea that Ranganathan conceived of that was most interesting, in my opinion, was using a syndetic structure involving colons to connect facets and build a context for inquiry -- the colon classification system. Why does this seem interesting? Because it offers a flexible syndetic architecture for building meaning out of attributes -- more flexible than a hierarchical approach to taxonomy creation. But just as systems using controlled vocabularies and indexing terms can co-exist with hierarchical taxonomies, there is no reason to suggest that you need to choose some sort of facet-based system over a hierarchical system or that one is superior over the other. Multiple modes of classification should work together in concert to allow for multiple paths to access. -Michael
COMMENT #4 My point is that you *so rarely* see faceted classifications (at least in consumer-facing information retrieval), which I find disappointing, because I think it is so useful. One reason you might not see them much is that it seems to be a hard concept for many to grasp--talking to the Argus folks who run the Synonyms and Taxonomies talk, they mention how the students always have trouble with facets. My main point is that by marrying faceted classification with good user centered design, we can create interfaces to information far more powerful and accessible than what is typically offered.
COMMENT #5 I mean, with any classification system you always have to add new categories or labels. With a hierarchical system, when you add a new concept you're done--it fits neatly into your scheme and no other part of the classification is affected. But with a faceted classification, when you add a new facet you will have to go over every single other item and see if it belongs to that facet. This doesn't seem feasible for large collections where humans would do the classification. And the alternatives--a static collection of facets or inconsistently applied facets--seem pretty bad. Maybe faceted classifications can only work when computers can do the classifying.
COMMENT #6 The authors describe some of the "whys" and "hows" of ontologogy development. Coincidentally, the primary example is a faceted classification system for wine. I hope others find this helpful as well. > seems to be a hard concept for many to grasp -- Maybe if they just called them "attributes" it wouldn't be so confusing. Todd [1] http://protege.stanford.edu/publications/ontology_development/ontology101.html COMMENT #7 As for maintenance, it's a matter of obtaining a thesaurus management type tool. I'll dig around for more resource to share here.
COMMENT #8 I thought the whole point of interaction design or user experience methodology (whatever you want to call it) was to devise and engage in a process that helps us solve problems on behalf of users, customers, readers (whatever you want to call them). So to ask the question, "How do user experience methods lead to innovation?", means that we're not understanding what innovation is and how we actually innovate. To my mind, innovation and creativity are radical and effective changes to something understood deeply, not the ability to make something out of nothing. Conventional wisdom about creativity has led to notions that only artists and designers are creative or that by "loosening up" people can become creative (as if creativity or the ability to innovate are things that individuals innately have or don't have). Conventional wisdom also often assumes that somehow creative solutions magically appear out of thin air and that any form of critical thinking is anathema to creativity and innovation. I say nay nay. First of all, innovation is a class of activity, not a characteristic of people. Second, creativity and innovation may certainly include accidental discoveries, but there certainly ain't no unintentional creativity. And last, innovation changes the systems that give things/objects/processes meaning. So the value of innovation is not creativity for creativity's sake, but in what innovation can provide in terms of solutions, differentiating advantage and other things like this. In other words, having lots of great ideas alone doesn't get you much of anywhere. What's required is the context in which those ideas become relevant and valuable. And one way (among many) that we can understand the context of something very deeply is to engage in research and definition processes espoused by user centered design: qualitative studies, ethnography, contextual inquiry, heuristic/expert reviews, task analysis, usability studies, etc. 2) Re: faceted classification: as Alex and Peter have stated: nothing new or innovative, but a great method for making classifcation schemes more accessible and relevant to the audiences who sift through and utilize the information-- or as our development teams used to say to clients, "great for creating affinity or niche marketing categories". It's key to understand Alex's point that facet-based classifcation schemes usually don't work well standing alone-- they require some other supportive structure (such as a hierarchy) in order to be useful and intelligble.
COMMENT #9 Per Todd's comment, I prefer the term 'Ontology'. Just as we don't say, 'let's do some parent-child classification' when we're creating a hierarchy, it may be more helpful to focus on the system, ontology, rather than a piece of the system, a facet.
COMMENT #10 Um... there's nothing faceted about the Best Cellars classification scheme. User-centered, absolutely; faceted, no. Also, the former philosophy major in me totally recoils at the use of the word 'ontology' in this context. Not that I have a better word to offer, mind you.
COMMENT #11 Most people (the Customers) do not think in hierarchical structures. They go looking for "something" that does "something". I still visit many websites that "demand" I know "their" nomencalture AND inventory control number for my "something". Well, kids, those folks don't get anything - except the sound of my wallet snaping shut as I "click-away" because nobody buys what he can't see or at least knows exists. "People covet what they see" People use and return to sites that are easy to use and useful to "THEIR" needs - regardless the sophistication of the "desinuhz" vocabulary or Flash skills. Remember, Play Nice with the Other Kids.
COMMENT #12 However, this can be problematic when you have many items that can fit into many classifications and sub-classifications. This can lead to classes and sub-classes that contain, what seems to users as, nearly every item imagined and they find the output overwhelming. One solution I have used is placing the items into the class and sub-class system and tie them to one dominant set of these parameters as well as to select as many sub-dominant class and sub-class options as the items properly fit into. This allows the users to select the classifications (and subs) that they believe will best produce their desired items. The result set will provide the items that are germane (by the classification system built based on user input and testing) and their sub-dominant associated items displayed in their germane classes and sub-classes. An example would be a system built for a tax site in the USA. User looking for information about Congress would select that classification and also select the option to view associated information. Their links returned would include Congress and its committees in the germane area. The query would also return, under their appropriate headings, non-profits following congress and newspapers/newsletters following congress on the taxation issues. From this point the users could click on the headings of the sub-dominant classification to view their full offerings. The prerequisite is to fully (as possible) understand the users/audience, which you are focussing your application/service. Cross-classification systems serve the user best, as they are more easily able to find information. From my experience tying the items to multiple class and sub-class buckets helps best cover a breadth of understanding in the user community. The down side to this is overwhelming the users with too much information, which can be mitigated, to some degree, with a dominant prioritization system.
COMMENT #13 I actually had the same thought in the shower this morning - that the wines were just put in new categories and there's nothing faceted about them. But I think Peter is implying how facets are at work here. A wine may have a 'sweet' value on the sweetness scale and a 'bubbly' value on the carbonation scale, so this wine could appear in several categories (and in the results of an 'advanced search' interface, an area where faceted categorization really shows its stuff). I agree the word Ontology is potentially problematic for those using the term in a philosophical context. This article attempts to bridge that gap: COMMENT #14 That's what I thought at first as well, but the site itself doesn't actually work that way. "Sweetness" and "bubblyness" aren't characteristics of all wines -- they're mutually exclusive categories into which wines are assigned. At least, that's the way the site is organized.
COMMENT #15 They feature this facet in context with others here: COMMENT #16 So now we're getting into more philosophical territory. If you have facets, but don't reveal them to the user, do you have a faceted classification scheme? My inclination is to say no. The whole point of doing faceted classification is to provide multiple access paths to the same content. If you close off every route but one, you aren't reaping any of the benefits of faceted classification.
COMMENT #17 Part of information architecture should be planning for the future, creating systems rather than simple classifications. Once again that's why I like 'Ontology' - it results in systems.
COMMENT #18 COMMENT #19 re: the relationship between faceted classification and ontologies, weiderhold [2] tends to organize things along a continuum of "increasing knowledge" [3]: symbol tables -> lexicons -> taxonomies -> db schemas -> data dictionaries -> object libraries -> domain models -> ontologies i see a large degree of overlap between the traditional ontological frame/class and slot/facet formalisms with "faceted classification" formalisms. while some may cringe at the baggage that gets associated with the ontology crowd and their association with expert systems and knowledge bases (see cyc [4]), it's important to distinguish between using ontologies for knowledge representation and information retrieval. "A Framework for Understanding and Classifying Ontology Applications"[5] illustrates several application scenarios that highlight the role of ontologies in support of information retrieval. [1]http://www.ukoln.ac.uk/metadata/desire/classification/ COMMENT #20 COMMENT #21 ontological formalisms will support this and more e.g. negation, conjunction, disjunction, recursion, relations, multiple inheritance, multi-valued slots, number restrictions on roles life science is one field where there is a load of work being done with ontology-oriented information systems, primarily because the conceptual model is exceedingly complex and the number of ways that users would like to interact with the data is even more complex. "An Evaluation of Ontology Exchange Languages for Bioinformatics as one might surmise, there is a degree of overlap, influence and inspiration between the ontology and semantic web [a la tim berners lee [3]] crowds. see, "Creating Semantic Web Contents with Protege-2000" [4]. protege is nice because it attempts to allow developers to create an ontology in one formalism like rdf and save it to another, e.g. OIL. if you haven't had enough yet, hours and hours of enjoyment can be found by perusing "Papers on Ontological Foundations of Conceptual Modelling and Knowledge Engineering" [5]. of particular relevance - "Supporting Ontological Analysis of Taxonomic Relationships." [6] [1]http://www.oasis-open.org/cover/OntologyExchange.html COMMENT #22 I agree that it is quite to often difficult to find a wine based on some of the current classification. What I find not so good with Cellars wine is that as a user, I need to remember what each of the term means (from the splash page). This ads burden on the user. Overall, they did a good job; the goal should be to accommodate all (wine novice, wine experts, etc.). Yves
COMMENT #23 Yeah, but these facet terms are pretty easy to grasp and self-describing, they're not exactly esoteric or hard to remember. That seems like one of the benefits of facets, that they offer access based on likely and common mental models of how stuff is categorized. I don't see the facets on Epicurious, has it changed? I do notice that they're starting to introduce other ways of offering up recipies: "Search Spy" and "most talked about" very like Amazon's ways of getting you to related or popular books.
COMMENT #24 COMMENT #25 For those of you who have not yet discovered it, there is a fully developed (although not yet fully published) faceted scheme, namely the Bliss Bibliographic Classification (2nd ed.), known as BC2. Its Editorial Committee is always pleased to hear from anyone who wishes to contribute to the continuing development of the scheme, regardless of subject area. For further details, please see the Bliss Classification Association homepage: COMMENT #26 tess l COMMENT #27 Here at the University College London we are currently working on the project that aims to apply facet analytical theory in building vocabulary for humanities portal (AHDS - www.ahds.ac.uk and Humbul - www.humbul.ac.uk). Our project site is www.ucl.ac.uk/fatks/ in case some of you may find interesting to see what we are working on here in the UK. Regards Aida Slavic COMMENT #28 BPAllen's search/browse of the same set of recipes is better because it is dynamic. The facets (I prefer "attributes") are presented to give you a notion of what recipes are available. Then the system filters results progressively based on the attributes chosen by the user. Select Italian cuisine, and you see how many of that subset of recipes have another attribute you're interested in (say, appetizer vs. main dish or dessert). Add one more attribute, say, pork, and you've narrowed your search down to a small, targeted list. If you change your mind and want to swap out pork for chicken, that's easy too. http://w3.bpallen.com:8080/spout/search.html?customer=demo&model=epidemo
COMMENT #29 COMMENT #30 COMMENT #31 Since 1999 the research and development has branched out into delivering systems rather than prototypes via View-Based Systems Ltd - based in the Yorkshire Technology Park situated near to Huddersfield. Apologies to those people who have been trying to follow links to papers on facetted classification and View-Based Searching - Huddersfield University relocated the research materials after I left! I have now put the relevant material onto the www.view-based-systems.com site. Publications and Presentations History Extracts with links are provided here. Individuals and organizations seeking collaboration please contact steve.pollitt@view-based-systems.com A S Pollitt, M P Smith and P A J Braekevelt (1996) View-based searching systems: A new paradigm for information retrieval based on faceted classification and indexing using mutually constraining knowledge-based rules Information Retrieval and Human Computer Interaction. Proceedings of the Joint workshop of the Information Retrieval and Human Computer Interaction Specialist Groups of the British Computer Society. pp 73-77, Johnson C. & Dunlop M (Eds) GIST Technical Report G96-2, Glasgow University, 17th September 1996. A S Pollitt (1997) The key role of classification and indexing in view-based searching IFLA '97 Copenhagen Aug 31 - Sept 3 1997 63rd IFLA General Conference Booklet 4, Section on Classification and Indexing Session 95 Paper 009-CLASS-1-E Note: A subset of 600,000 records from EMBASE provided the database held on an ADABAS (inverted file) database management system. Visual Basic provided the interface development language. A Sun Sparc Server 2 provided the searching hardware. Retrieval speed was acceptable (as evidenced from the screen cam demonstrations downloadable from View-Based Systems). The usability was demonstrated in an experiment with end users at the Medical Library in the University of Leeds. Unfortunately the retrieval performance was not compared with the current retrieval systems as they used MEDLINE. A S Pollitt (1998) Faceted-Classification as Pre-Co-ordinated Subject Indexing: Multi-dimensional Searching for OPAC Users, Knowledge Organisation 98, Norwegian Classification and Indexing Society, Oslo College, 5-6 May 1998 A S Pollitt and A Tinker (2000) Navigating N-Dimensional Information Space with Data and Documents through View-Based Searching, in Proceedings of 22nd Annual Colloquium on Information Retrieval Research 5th-7th April 2000, Sidney Sussex College, Cambridge pp 232-241 (PowerPoint Presentation self-extracting zipped file) Dr A Steven Pollitt COMMENT #32
|
